Install
openclaw skills install @tianzhiceng297-boop/minimal-edit对既有文本做最小化局部修改,保持语气与篇幅,不展开、不强调、不标注改动。Surgeon-precise, tone-preserving edits to existing text.
openclaw skills install @tianzhiceng297-boop/minimal-editTreat every request as a local edit unless the user explicitly asks for a full rewrite. Change exactly the semantic unit in scope and preserve everything else: sentence rhythm, length, emphasis, structure, tone, facts, examples, and formatting.
Replace the semantic claim, not the paragraph around it. Keep the same sentence structure, level of detail, and length. If the original is one clause, the replacement stays one clause. Do not add caveats, examples, reasoning, or transition sentences unless the user asks for them.
If the user says to recommend something implicitly, keep the recommendation implicit. Do not turn it into an explicit conclusion or a list of disclaimers.
Remove the target text. Fix only the connectors needed to keep the sentence grammatical and the surrounding flow intact. Do not add a replacement sentence that summarizes what was deleted. Do not add "总之", "因此", or a new conclusion to fill the gap.
Remove or reduce the visual and rhetorical weight: drop bold, headings, callouts, repeated examples, and superlatives if they are the reason the passage stands out. Keep the factual content, but state it at the same level as neighboring sentences.
Change only the incorrect fact or expression. Preserve the rest of the sentence exactly, including its punctuation and surrounding clauses.
Remove formulaic AI-sounding phrases and mechanical structure. Keep facts, logic, and the user's point. Do not replace one AI phrase with another AI phrase, and do not add new framing or summary sentences.
Common Chinese candidates include 赋能, 抓手, 闭环, 颗粒度, 场景化, 底层逻辑, 战略协同, 深度绑定, 一站式, 全方位, 多维度, 系统性, 组合拳, 矩阵, 拉通, 对齐, 降本增效, and 提质增效. Common English candidates include "it is worth noting", "in conclusion", "leverage", "synergy", "holistic", "ecosystem", and "seamless". Treat these as manual-review candidates, not automatic proof of a problem.
When a request combines operations, such as "delete this sentence and soften the tone", split it into atomic operations first. Apply content changes first, then tone, then emphasis and formatting. Keep one invariant set across the whole batch: the final result must be the smallest change that satisfies every part. If two parts conflict, choose the lighter interpretation and keep the final text minimal. Audit against every requested operation, not just the first one.
Return the clean revised text only. If the user asks why something changed, explain in a separate short note outside the deliverable, never with inline markers in the final text.
python scripts/audit_edit.py --before <original> --after <edited>python scripts/audit_edit.py --before-text "<original>" --after-text "<revised>"--must-remove "<phrase>" for each deleted target.
Fix or manually review every warning before delivering. Marker warnings are candidates, not proof that the text is clean.When the task feels ambiguous or you need concrete before/after patterns, read references/examples.md and match the closest case.